holmesgpt

HolmesGPT is an open-source AI agent for investigating production incidents and finding root causes, and a Cloud Native Computing Foundation (CNCF) sandbox project. Given an alert or a question, it runs an agentic loop that queries live observability data across your stack (Kubernetes, VMs, cloud providers, databases and SaaS platforms) and explains what went wrong.

It suits SRE, platform and on-call teams who want faster triage. Built-in toolsets connect to Prometheus, Grafana, Datadog, Kubernetes and many other sources, plus any REST API; alert integrations fetch alerts from AlertManager, PagerDuty, OpsGenie or Jira and write findings back. It is engineered for large data: server-side filtering, output transformers and memory limits keep huge payloads out of the model's context and avoid out-of-memory crashes. The newer Operator mode runs in Kubernetes and performs deployment verification and scheduled health checks in the background, so problems can be caught before a human notices them. It works with OpenAI, Anthropic, Azure, Bedrock, Gemini and other providers, and does not require Kubernetes.

Getting started: follow the installation documentation for the CLI or your preferred deployment method, connect a model provider and the data sources you use, then ask it to investigate an alert or a symptom.